# numaperf

> NUMA-first runtime for latency-critical Rust applications. Explicit control over memory placement, thread pinning, and work scheduling on NUMA systems. Maintained by Skelf Research. MIT licensed.

numaperf provides topology discovery, RAII thread pinning, explicit memory placement policies (Bind, Preferred, Interleave, Local), per-node work scheduling, per-node sharded data structures, device locality, and locality observability. It targets Linux x86_64 and aarch64 with graceful degradation on macOS. MSRV 1.70.

## Primary surfaces

- [Overview](https://numaperf.skelfresearch.com/): hero, features, code example, platform support, use cases.
- [About](https://numaperf.skelfresearch.com/about): scope, philosophy, status, license.
- [Notes (blog)](https://numaperf.skelfresearch.com/blog): field notes on NUMA, tail latency, and Rust runtime gaps.
- [Compare](https://numaperf.skelfresearch.com/compare/): structural comparisons against adjacent projects.

## Notes

- [When NUMA actually matters: a recipe for the p99 hunter](https://numaperf.skelfresearch.com/blog/when-numa-actually-matters/)
- [Pinning vs scheduling: where Rust runtimes leave perf on the table](https://numaperf.skelfresearch.com/blog/pinning-vs-scheduling/)
- [Reading numastat without lying to yourself](https://numaperf.skelfresearch.com/blog/reading-numastat-without-lying/)

## Comparisons

- [numaperf vs. Tokio](https://numaperf.skelfresearch.com/compare/tokio/)
- [numaperf vs. Glommio](https://numaperf.skelfresearch.com/compare/glommio/)

## Off-site canonical resources

- Documentation: https://docs.skelfresearch.com/numaperf/
- Source: https://github.com/Skelf-Research/numaperf
- Crate: https://crates.io/crates/numaperf
- Issue tracker: https://github.com/Skelf-Research/numaperf/issues

## Out of scope for this site

- No pricing, tokens, or sponsorship tiers — numaperf is MIT-licensed open source.
- No benchmark numbers fabricated for marketing. Where the project has not published a measurement, this site does not invent one.